cgroup-defs.h 27 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877
  1. /* SPDX-License-Identifier: GPL-2.0 */
  2. /*
  3. * linux/cgroup-defs.h - basic definitions for cgroup
  4. *
  5. * This file provides basic type and interface. Include this file directly
  6. * only if necessary to avoid cyclic dependencies.
  7. */
  8. #ifndef _LINUX_CGROUP_DEFS_H
  9. #define _LINUX_CGROUP_DEFS_H
  10. #include <linux/limits.h>
  11. #include <linux/list.h>
  12. #include <linux/idr.h>
  13. #include <linux/wait.h>
  14. #include <linux/mutex.h>
  15. #include <linux/rcupdate.h>
  16. #include <linux/refcount.h>
  17. #include <linux/percpu-refcount.h>
  18. #include <linux/percpu-rwsem.h>
  19. #include <linux/u64_stats_sync.h>
  20. #include <linux/workqueue.h>
  21. #include <linux/bpf-cgroup.h>
  22. #include <linux/psi_types.h>
  23. #ifdef CONFIG_CGROUPS
  24. struct cgroup;
  25. struct cgroup_root;
  26. struct cgroup_subsys;
  27. struct cgroup_taskset;
  28. struct kernfs_node;
  29. struct kernfs_ops;
  30. struct kernfs_open_file;
  31. struct seq_file;
  32. struct poll_table_struct;
  33. #define MAX_CGROUP_TYPE_NAMELEN 32
  34. #define MAX_CGROUP_ROOT_NAMELEN 64
  35. #define MAX_CFTYPE_NAME 64
  36. /* define the enumeration of all cgroup subsystems */
  37. #define SUBSYS(_x) _x ## _cgrp_id,
  38. enum cgroup_subsys_id {
  39. #include <linux/cgroup_subsys.h>
  40. CGROUP_SUBSYS_COUNT,
  41. };
  42. #undef SUBSYS
  43. /* bits in struct cgroup_subsys_state flags field */
  44. enum {
  45. CSS_NO_REF = (1 << 0), /* no reference counting for this css */
  46. CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */
  47. CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */
  48. CSS_VISIBLE = (1 << 3), /* css is visible to userland */
  49. CSS_DYING = (1 << 4), /* css is dying */
  50. };
  51. /* bits in struct cgroup flags field */
  52. enum {
  53. /* Control Group requires release notifications to userspace */
  54. CGRP_NOTIFY_ON_RELEASE,
  55. /*
  56. * Clone the parent's configuration when creating a new child
  57. * cpuset cgroup. For historical reasons, this option can be
  58. * specified at mount time and thus is implemented here.
  59. */
  60. CGRP_CPUSET_CLONE_CHILDREN,
  61. /* Control group has to be frozen. */
  62. CGRP_FREEZE,
  63. /* Cgroup is frozen. */
  64. CGRP_FROZEN,
  65. };
  66. /* cgroup_root->flags */
  67. enum {
  68. CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */
  69. CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */
  70. /*
  71. * Consider namespaces as delegation boundaries. If this flag is
  72. * set, controller specific interface files in a namespace root
  73. * aren't writeable from inside the namespace.
  74. */
  75. CGRP_ROOT_NS_DELEGATE = (1 << 3),
  76. /*
  77. * Enable cpuset controller in v1 cgroup to use v2 behavior.
  78. */
  79. CGRP_ROOT_CPUSET_V2_MODE = (1 << 4),
  80. /*
  81. * Enable legacy local memory.events.
  82. */
  83. CGRP_ROOT_MEMORY_LOCAL_EVENTS = (1 << 5),
  84. /*
  85. * Enable recursive subtree protection
  86. */
  87. CGRP_ROOT_MEMORY_RECURSIVE_PROT = (1 << 6),
  88. };
  89. /* cftype->flags */
  90. enum {
  91. CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */
  92. CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */
  93. CFTYPE_NS_DELEGATABLE = (1 << 2), /* writeable beyond delegation boundaries */
  94. CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */
  95. CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */
  96. CFTYPE_DEBUG = (1 << 5), /* create when cgroup_debug */
  97. CFTYPE_PRESSURE = (1 << 6), /* only if pressure feature is enabled */
  98. /* internal flags, do not use outside cgroup core proper */
  99. __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */
  100. __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */
  101. };
  102. /*
  103. * cgroup_file is the handle for a file instance created in a cgroup which
  104. * is used, for example, to generate file changed notifications. This can
  105. * be obtained by setting cftype->file_offset.
  106. */
  107. struct cgroup_file {
  108. /* do not access any fields from outside cgroup core */
  109. struct kernfs_node *kn;
  110. unsigned long notified_at;
  111. struct timer_list notify_timer;
  112. };
  113. /*
  114. * Per-subsystem/per-cgroup state maintained by the system. This is the
  115. * fundamental structural building block that controllers deal with.
  116. *
  117. * Fields marked with "PI:" are public and immutable and may be accessed
  118. * directly without synchronization.
  119. */
  120. struct cgroup_subsys_state {
  121. /* PI: the cgroup that this css is attached to */
  122. struct cgroup *cgroup;
  123. /* PI: the cgroup subsystem that this css is attached to */
  124. struct cgroup_subsys *ss;
  125. /* reference count - access via css_[try]get() and css_put() */
  126. struct percpu_ref refcnt;
  127. /* siblings list anchored at the parent's ->children */
  128. struct list_head sibling;
  129. struct list_head children;
  130. /* flush target list anchored at cgrp->rstat_css_list */
  131. struct list_head rstat_css_node;
  132. /*
  133. * PI: Subsys-unique ID. 0 is unused and root is always 1. The
  134. * matching css can be looked up using css_from_id().
  135. */
  136. int id;
  137. unsigned int flags;
  138. /*
  139. * Monotonically increasing unique serial number which defines a
  140. * uniform order among all csses. It's guaranteed that all
  141. * ->children lists are in the ascending order of ->serial_nr and
  142. * used to allow interrupting and resuming iterations.
  143. */
  144. u64 serial_nr;
  145. /*
  146. * Incremented by online self and children. Used to guarantee that
  147. * parents are not offlined before their children.
  148. */
  149. atomic_t online_cnt;
  150. /* percpu_ref killing and RCU release */
  151. struct work_struct destroy_work;
  152. struct rcu_work destroy_rwork;
  153. /*
  154. * PI: the parent css. Placed here for cache proximity to following
  155. * fields of the containing structure.
  156. */
  157. struct cgroup_subsys_state *parent;
  158. };
  159. /*
  160. * A css_set is a structure holding pointers to a set of
  161. * cgroup_subsys_state objects. This saves space in the task struct
  162. * object and speeds up fork()/exit(), since a single inc/dec and a
  163. * list_add()/del() can bump the reference count on the entire cgroup
  164. * set for a task.
  165. */
  166. struct css_set {
  167. /*
  168. * Set of subsystem states, one for each subsystem. This array is
  169. * immutable after creation apart from the init_css_set during
  170. * subsystem registration (at boot time).
  171. */
  172. struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT];
  173. /* reference count */
  174. refcount_t refcount;
  175. /*
  176. * For a domain cgroup, the following points to self. If threaded,
  177. * to the matching cset of the nearest domain ancestor. The
  178. * dom_cset provides access to the domain cgroup and its csses to
  179. * which domain level resource consumptions should be charged.
  180. */
  181. struct css_set *dom_cset;
  182. /* the default cgroup associated with this css_set */
  183. struct cgroup *dfl_cgrp;
  184. /* internal task count, protected by css_set_lock */
  185. int nr_tasks;
  186. /*
  187. * Lists running through all tasks using this cgroup group.
  188. * mg_tasks lists tasks which belong to this cset but are in the
  189. * process of being migrated out or in. Protected by
  190. * css_set_rwsem, but, during migration, once tasks are moved to
  191. * mg_tasks, it can be read safely while holding cgroup_mutex.
  192. */
  193. struct list_head tasks;
  194. struct list_head mg_tasks;
  195. struct list_head dying_tasks;
  196. /* all css_task_iters currently walking this cset */
  197. struct list_head task_iters;
  198. /*
  199. * On the default hierarhcy, ->subsys[ssid] may point to a css
  200. * attached to an ancestor instead of the cgroup this css_set is
  201. * associated with. The following node is anchored at
  202. * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to
  203. * iterate through all css's attached to a given cgroup.
  204. */
  205. struct list_head e_cset_node[CGROUP_SUBSYS_COUNT];
  206. /* all threaded csets whose ->dom_cset points to this cset */
  207. struct list_head threaded_csets;
  208. struct list_head threaded_csets_node;
  209. /*
  210. * List running through all cgroup groups in the same hash
  211. * slot. Protected by css_set_lock
  212. */
  213. struct hlist_node hlist;
  214. /*
  215. * List of cgrp_cset_links pointing at cgroups referenced from this
  216. * css_set. Protected by css_set_lock.
  217. */
  218. struct list_head cgrp_links;
  219. /*
  220. * List of csets participating in the on-going migration either as
  221. * source or destination. Protected by cgroup_mutex.
  222. */
  223. struct list_head mg_preload_node;
  224. struct list_head mg_node;
  225. /*
  226. * If this cset is acting as the source of migration the following
  227. * two fields are set. mg_src_cgrp and mg_dst_cgrp are
  228. * respectively the source and destination cgroups of the on-going
  229. * migration. mg_dst_cset is the destination cset the target tasks
  230. * on this cset should be migrated to. Protected by cgroup_mutex.
  231. */
  232. struct cgroup *mg_src_cgrp;
  233. struct cgroup *mg_dst_cgrp;
  234. struct css_set *mg_dst_cset;
  235. /* dead and being drained, ignore for migration */
  236. bool dead;
  237. /* For RCU-protected deletion */
  238. struct rcu_head rcu_head;
  239. };
  240. struct cgroup_base_stat {
  241. struct task_cputime cputime;
  242. };
  243. /*
  244. * rstat - cgroup scalable recursive statistics. Accounting is done
  245. * per-cpu in cgroup_rstat_cpu which is then lazily propagated up the
  246. * hierarchy on reads.
  247. *
  248. * When a stat gets updated, the cgroup_rstat_cpu and its ancestors are
  249. * linked into the updated tree. On the following read, propagation only
  250. * considers and consumes the updated tree. This makes reading O(the
  251. * number of descendants which have been active since last read) instead of
  252. * O(the total number of descendants).
  253. *
  254. * This is important because there can be a lot of (draining) cgroups which
  255. * aren't active and stat may be read frequently. The combination can
  256. * become very expensive. By propagating selectively, increasing reading
  257. * frequency decreases the cost of each read.
  258. *
  259. * This struct hosts both the fields which implement the above -
  260. * updated_children and updated_next - and the fields which track basic
  261. * resource statistics on top of it - bsync, bstat and last_bstat.
  262. */
  263. struct cgroup_rstat_cpu {
  264. /*
  265. * ->bsync protects ->bstat. These are the only fields which get
  266. * updated in the hot path.
  267. */
  268. struct u64_stats_sync bsync;
  269. struct cgroup_base_stat bstat;
  270. /*
  271. * Snapshots at the last reading. These are used to calculate the
  272. * deltas to propagate to the global counters.
  273. */
  274. struct cgroup_base_stat last_bstat;
  275. /*
  276. * Child cgroups with stat updates on this cpu since the last read
  277. * are linked on the parent's ->updated_children through
  278. * ->updated_next.
  279. *
  280. * In addition to being more compact, singly-linked list pointing
  281. * to the cgroup makes it unnecessary for each per-cpu struct to
  282. * point back to the associated cgroup.
  283. *
  284. * Protected by per-cpu cgroup_rstat_cpu_lock.
  285. */
  286. struct cgroup *updated_children; /* terminated by self cgroup */
  287. struct cgroup *updated_next; /* NULL iff not on the list */
  288. };
  289. struct cgroup_freezer_state {
  290. /* Should the cgroup and its descendants be frozen. */
  291. bool freeze;
  292. /* Should the cgroup actually be frozen? */
  293. int e_freeze;
  294. /* Fields below are protected by css_set_lock */
  295. /* Number of frozen descendant cgroups */
  296. int nr_frozen_descendants;
  297. /*
  298. * Number of tasks, which are counted as frozen:
  299. * frozen, SIGSTOPped, and PTRACEd.
  300. */
  301. int nr_frozen_tasks;
  302. };
  303. struct cgroup {
  304. /* self css with NULL ->ss, points back to this cgroup */
  305. struct cgroup_subsys_state self;
  306. unsigned long flags; /* "unsigned long" so bitops work */
  307. /*
  308. * The depth this cgroup is at. The root is at depth zero and each
  309. * step down the hierarchy increments the level. This along with
  310. * ancestor_ids[] can determine whether a given cgroup is a
  311. * descendant of another without traversing the hierarchy.
  312. */
  313. int level;
  314. /* Maximum allowed descent tree depth */
  315. int max_depth;
  316. /*
  317. * Keep track of total numbers of visible and dying descent cgroups.
  318. * Dying cgroups are cgroups which were deleted by a user,
  319. * but are still existing because someone else is holding a reference.
  320. * max_descendants is a maximum allowed number of descent cgroups.
  321. *
  322. * nr_descendants and nr_dying_descendants are protected
  323. * by cgroup_mutex and css_set_lock. It's fine to read them holding
  324. * any of cgroup_mutex and css_set_lock; for writing both locks
  325. * should be held.
  326. */
  327. int nr_descendants;
  328. int nr_dying_descendants;
  329. int max_descendants;
  330. /*
  331. * Each non-empty css_set associated with this cgroup contributes
  332. * one to nr_populated_csets. The counter is zero iff this cgroup
  333. * doesn't have any tasks.
  334. *
  335. * All children which have non-zero nr_populated_csets and/or
  336. * nr_populated_children of their own contribute one to either
  337. * nr_populated_domain_children or nr_populated_threaded_children
  338. * depending on their type. Each counter is zero iff all cgroups
  339. * of the type in the subtree proper don't have any tasks.
  340. */
  341. int nr_populated_csets;
  342. int nr_populated_domain_children;
  343. int nr_populated_threaded_children;
  344. int nr_threaded_children; /* # of live threaded child cgroups */
  345. struct kernfs_node *kn; /* cgroup kernfs entry */
  346. struct cgroup_file procs_file; /* handle for "cgroup.procs" */
  347. struct cgroup_file events_file; /* handle for "cgroup.events" */
  348. /*
  349. * The bitmask of subsystems enabled on the child cgroups.
  350. * ->subtree_control is the one configured through
  351. * "cgroup.subtree_control" while ->child_ss_mask is the effective
  352. * one which may have more subsystems enabled. Controller knobs
  353. * are made available iff it's enabled in ->subtree_control.
  354. */
  355. u16 subtree_control;
  356. u16 subtree_ss_mask;
  357. u16 old_subtree_control;
  358. u16 old_subtree_ss_mask;
  359. /* Private pointers for each registered subsystem */
  360. struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT];
  361. struct cgroup_root *root;
  362. /*
  363. * List of cgrp_cset_links pointing at css_sets with tasks in this
  364. * cgroup. Protected by css_set_lock.
  365. */
  366. struct list_head cset_links;
  367. /*
  368. * On the default hierarchy, a css_set for a cgroup with some
  369. * susbsys disabled will point to css's which are associated with
  370. * the closest ancestor which has the subsys enabled. The
  371. * following lists all css_sets which point to this cgroup's css
  372. * for the given subsystem.
  373. */
  374. struct list_head e_csets[CGROUP_SUBSYS_COUNT];
  375. /*
  376. * If !threaded, self. If threaded, it points to the nearest
  377. * domain ancestor. Inside a threaded subtree, cgroups are exempt
  378. * from process granularity and no-internal-task constraint.
  379. * Domain level resource consumptions which aren't tied to a
  380. * specific task are charged to the dom_cgrp.
  381. */
  382. struct cgroup *dom_cgrp;
  383. struct cgroup *old_dom_cgrp; /* used while enabling threaded */
  384. /* per-cpu recursive resource statistics */
  385. struct cgroup_rstat_cpu __percpu *rstat_cpu;
  386. struct list_head rstat_css_list;
  387. /* cgroup basic resource statistics */
  388. struct cgroup_base_stat last_bstat;
  389. struct cgroup_base_stat bstat;
  390. struct prev_cputime prev_cputime; /* for printing out cputime */
  391. /*
  392. * list of pidlists, up to two for each namespace (one for procs, one
  393. * for tasks); created on demand.
  394. */
  395. struct list_head pidlists;
  396. struct mutex pidlist_mutex;
  397. /* used to wait for offlining of csses */
  398. wait_queue_head_t offline_waitq;
  399. /* used to schedule release agent */
  400. struct work_struct release_agent_work;
  401. /* used to track pressure stalls */
  402. struct psi_group psi;
  403. /* used to store eBPF programs */
  404. struct cgroup_bpf bpf;
  405. /* If there is block congestion on this cgroup. */
  406. atomic_t congestion_count;
  407. /* Used to store internal freezer state */
  408. struct cgroup_freezer_state freezer;
  409. /* ids of the ancestors at each level including self */
  410. u64 ancestor_ids[];
  411. };
  412. /*
  413. * A cgroup_root represents the root of a cgroup hierarchy, and may be
  414. * associated with a kernfs_root to form an active hierarchy. This is
  415. * internal to cgroup core. Don't access directly from controllers.
  416. */
  417. struct cgroup_root {
  418. struct kernfs_root *kf_root;
  419. /* The bitmask of subsystems attached to this hierarchy */
  420. unsigned int subsys_mask;
  421. /* Unique id for this hierarchy. */
  422. int hierarchy_id;
  423. /* The root cgroup. Root is destroyed on its release. */
  424. struct cgroup cgrp;
  425. /* for cgrp->ancestor_ids[0] */
  426. u64 cgrp_ancestor_id_storage;
  427. /* Number of cgroups in the hierarchy, used only for /proc/cgroups */
  428. atomic_t nr_cgrps;
  429. /* A list running through the active hierarchies */
  430. struct list_head root_list;
  431. /* Hierarchy-specific flags */
  432. unsigned int flags;
  433. /* The path to use for release notifications. */
  434. char release_agent_path[PATH_MAX];
  435. /* The name for this hierarchy - may be empty */
  436. char name[MAX_CGROUP_ROOT_NAMELEN];
  437. };
  438. /*
  439. * struct cftype: handler definitions for cgroup control files
  440. *
  441. * When reading/writing to a file:
  442. * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata
  443. * - the 'cftype' of the file is file->f_path.dentry->d_fsdata
  444. */
  445. struct cftype {
  446. /*
  447. * By convention, the name should begin with the name of the
  448. * subsystem, followed by a period. Zero length string indicates
  449. * end of cftype array.
  450. */
  451. char name[MAX_CFTYPE_NAME];
  452. unsigned long private;
  453. /*
  454. * The maximum length of string, excluding trailing nul, that can
  455. * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed.
  456. */
  457. size_t max_write_len;
  458. /* CFTYPE_* flags */
  459. unsigned int flags;
  460. /*
  461. * If non-zero, should contain the offset from the start of css to
  462. * a struct cgroup_file field. cgroup will record the handle of
  463. * the created file into it. The recorded handle can be used as
  464. * long as the containing css remains accessible.
  465. */
  466. unsigned int file_offset;
  467. /*
  468. * Fields used for internal bookkeeping. Initialized automatically
  469. * during registration.
  470. */
  471. struct cgroup_subsys *ss; /* NULL for cgroup core files */
  472. struct list_head node; /* anchored at ss->cfts */
  473. struct kernfs_ops *kf_ops;
  474. int (*open)(struct kernfs_open_file *of);
  475. void (*release)(struct kernfs_open_file *of);
  476. /*
  477. * read_u64() is a shortcut for the common case of returning a
  478. * single integer. Use it in place of read()
  479. */
  480. u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft);
  481. /*
  482. * read_s64() is a signed version of read_u64()
  483. */
  484. s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft);
  485. /* generic seq_file read interface */
  486. int (*seq_show)(struct seq_file *sf, void *v);
  487. /* optional ops, implement all or none */
  488. void *(*seq_start)(struct seq_file *sf, loff_t *ppos);
  489. void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos);
  490. void (*seq_stop)(struct seq_file *sf, void *v);
  491. /*
  492. * write_u64() is a shortcut for the common case of accepting
  493. * a single integer (as parsed by simple_strtoull) from
  494. * userspace. Use in place of write(); return 0 or error.
  495. */
  496. int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft,
  497. u64 val);
  498. /*
  499. * write_s64() is a signed version of write_u64()
  500. */
  501. int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft,
  502. s64 val);
  503. /*
  504. * write() is the generic write callback which maps directly to
  505. * kernfs write operation and overrides all other operations.
  506. * Maximum write size is determined by ->max_write_len. Use
  507. * of_css/cft() to access the associated css and cft.
  508. */
  509. ssize_t (*write)(struct kernfs_open_file *of,
  510. char *buf, size_t nbytes, loff_t off);
  511. __poll_t (*poll)(struct kernfs_open_file *of,
  512. struct poll_table_struct *pt);
  513. #ifdef CONFIG_DEBUG_LOCK_ALLOC
  514. struct lock_class_key lockdep_key;
  515. #endif
  516. };
  517. /*
  518. * Control Group subsystem type.
  519. * See Documentation/admin-guide/cgroup-v1/cgroups.rst for details
  520. */
  521. struct cgroup_subsys {
  522. struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css);
  523. int (*css_online)(struct cgroup_subsys_state *css);
  524. void (*css_offline)(struct cgroup_subsys_state *css);
  525. void (*css_released)(struct cgroup_subsys_state *css);
  526. void (*css_free)(struct cgroup_subsys_state *css);
  527. void (*css_reset)(struct cgroup_subsys_state *css);
  528. void (*css_rstat_flush)(struct cgroup_subsys_state *css, int cpu);
  529. int (*css_extra_stat_show)(struct seq_file *seq,
  530. struct cgroup_subsys_state *css);
  531. int (*can_attach)(struct cgroup_taskset *tset);
  532. void (*cancel_attach)(struct cgroup_taskset *tset);
  533. void (*attach)(struct cgroup_taskset *tset);
  534. void (*post_attach)(void);
  535. int (*can_fork)(struct task_struct *task,
  536. struct css_set *cset);
  537. void (*cancel_fork)(struct task_struct *task, struct css_set *cset);
  538. void (*fork)(struct task_struct *task);
  539. void (*exit)(struct task_struct *task);
  540. void (*release)(struct task_struct *task);
  541. void (*bind)(struct cgroup_subsys_state *root_css);
  542. bool early_init:1;
  543. /*
  544. * If %true, the controller, on the default hierarchy, doesn't show
  545. * up in "cgroup.controllers" or "cgroup.subtree_control", is
  546. * implicitly enabled on all cgroups on the default hierarchy, and
  547. * bypasses the "no internal process" constraint. This is for
  548. * utility type controllers which is transparent to userland.
  549. *
  550. * An implicit controller can be stolen from the default hierarchy
  551. * anytime and thus must be okay with offline csses from previous
  552. * hierarchies coexisting with csses for the current one.
  553. */
  554. bool implicit_on_dfl:1;
  555. /*
  556. * If %true, the controller, supports threaded mode on the default
  557. * hierarchy. In a threaded subtree, both process granularity and
  558. * no-internal-process constraint are ignored and a threaded
  559. * controllers should be able to handle that.
  560. *
  561. * Note that as an implicit controller is automatically enabled on
  562. * all cgroups on the default hierarchy, it should also be
  563. * threaded. implicit && !threaded is not supported.
  564. */
  565. bool threaded:1;
  566. /*
  567. * If %false, this subsystem is properly hierarchical -
  568. * configuration, resource accounting and restriction on a parent
  569. * cgroup cover those of its children. If %true, hierarchy support
  570. * is broken in some ways - some subsystems ignore hierarchy
  571. * completely while others are only implemented half-way.
  572. *
  573. * It's now disallowed to create nested cgroups if the subsystem is
  574. * broken and cgroup core will emit a warning message on such
  575. * cases. Eventually, all subsystems will be made properly
  576. * hierarchical and this will go away.
  577. */
  578. bool broken_hierarchy:1;
  579. bool warned_broken_hierarchy:1;
  580. /* the following two fields are initialized automtically during boot */
  581. int id;
  582. const char *name;
  583. /* optional, initialized automatically during boot if not set */
  584. const char *legacy_name;
  585. /* link to parent, protected by cgroup_lock() */
  586. struct cgroup_root *root;
  587. /* idr for css->id */
  588. struct idr css_idr;
  589. /*
  590. * List of cftypes. Each entry is the first entry of an array
  591. * terminated by zero length name.
  592. */
  593. struct list_head cfts;
  594. /*
  595. * Base cftypes which are automatically registered. The two can
  596. * point to the same array.
  597. */
  598. struct cftype *dfl_cftypes; /* for the default hierarchy */
  599. struct cftype *legacy_cftypes; /* for the legacy hierarchies */
  600. /*
  601. * A subsystem may depend on other subsystems. When such subsystem
  602. * is enabled on a cgroup, the depended-upon subsystems are enabled
  603. * together if available. Subsystems enabled due to dependency are
  604. * not visible to userland until explicitly enabled. The following
  605. * specifies the mask of subsystems that this one depends on.
  606. */
  607. unsigned int depends_on;
  608. };
  609. extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem;
  610. /**
  611. * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups
  612. * @tsk: target task
  613. *
  614. * Allows cgroup operations to synchronize against threadgroup changes
  615. * using a percpu_rw_semaphore.
  616. */
  617. static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
  618. {
  619. percpu_down_read(&cgroup_threadgroup_rwsem);
  620. }
  621. /**
  622. * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups
  623. * @tsk: target task
  624. *
  625. * Counterpart of cgroup_threadcgroup_change_begin().
  626. */
  627. static inline void cgroup_threadgroup_change_end(struct task_struct *tsk)
  628. {
  629. percpu_up_read(&cgroup_threadgroup_rwsem);
  630. }
  631. #else /* CONFIG_CGROUPS */
  632. #define CGROUP_SUBSYS_COUNT 0
  633. static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk)
  634. {
  635. might_sleep();
  636. }
  637. static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {}
  638. #endif /* CONFIG_CGROUPS */
  639. #ifdef CONFIG_SOCK_CGROUP_DATA
  640. /*
  641. * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains
  642. * per-socket cgroup information except for memcg association.
  643. *
  644. * On legacy hierarchies, net_prio and net_cls controllers directly set
  645. * attributes on each sock which can then be tested by the network layer.
  646. * On the default hierarchy, each sock is associated with the cgroup it was
  647. * created in and the networking layer can match the cgroup directly.
  648. *
  649. * To avoid carrying all three cgroup related fields separately in sock,
  650. * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer.
  651. * On boot, sock_cgroup_data records the cgroup that the sock was created
  652. * in so that cgroup2 matches can be made; however, once either net_prio or
  653. * net_cls starts being used, the area is overriden to carry prioidx and/or
  654. * classid. The two modes are distinguished by whether the lowest bit is
  655. * set. Clear bit indicates cgroup pointer while set bit prioidx and
  656. * classid.
  657. *
  658. * While userland may start using net_prio or net_cls at any time, once
  659. * either is used, cgroup2 matching no longer works. There is no reason to
  660. * mix the two and this is in line with how legacy and v2 compatibility is
  661. * handled. On mode switch, cgroup references which are already being
  662. * pointed to by socks may be leaked. While this can be remedied by adding
  663. * synchronization around sock_cgroup_data, given that the number of leaked
  664. * cgroups is bound and highly unlikely to be high, this seems to be the
  665. * better trade-off.
  666. */
  667. struct sock_cgroup_data {
  668. union {
  669. #ifdef __LITTLE_ENDIAN
  670. struct {
  671. u8 is_data : 1;
  672. u8 no_refcnt : 1;
  673. u8 unused : 6;
  674. u8 padding;
  675. u16 prioidx;
  676. u32 classid;
  677. } __packed;
  678. #else
  679. struct {
  680. u32 classid;
  681. u16 prioidx;
  682. u8 padding;
  683. u8 unused : 6;
  684. u8 no_refcnt : 1;
  685. u8 is_data : 1;
  686. } __packed;
  687. #endif
  688. u64 val;
  689. };
  690. };
  691. /*
  692. * There's a theoretical window where the following accessors race with
  693. * updaters and return part of the previous pointer as the prioidx or
  694. * classid. Such races are short-lived and the result isn't critical.
  695. */
  696. static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd)
  697. {
  698. /* fallback to 1 which is always the ID of the root cgroup */
  699. return (skcd->is_data & 1) ? skcd->prioidx : 1;
  700. }
  701. static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd)
  702. {
  703. /* fallback to 0 which is the unconfigured default classid */
  704. return (skcd->is_data & 1) ? skcd->classid : 0;
  705. }
  706. /*
  707. * If invoked concurrently, the updaters may clobber each other. The
  708. * caller is responsible for synchronization.
  709. */
  710. static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd,
  711. u16 prioidx)
  712. {
  713. struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
  714. if (sock_cgroup_prioidx(&skcd_buf) == prioidx)
  715. return;
  716. if (!(skcd_buf.is_data & 1)) {
  717. skcd_buf.val = 0;
  718. skcd_buf.is_data = 1;
  719. }
  720. skcd_buf.prioidx = prioidx;
  721. WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
  722. }
  723. static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd,
  724. u32 classid)
  725. {
  726. struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }};
  727. if (sock_cgroup_classid(&skcd_buf) == classid)
  728. return;
  729. if (!(skcd_buf.is_data & 1)) {
  730. skcd_buf.val = 0;
  731. skcd_buf.is_data = 1;
  732. }
  733. skcd_buf.classid = classid;
  734. WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */
  735. }
  736. #else /* CONFIG_SOCK_CGROUP_DATA */
  737. struct sock_cgroup_data {
  738. };
  739. #endif /* CONFIG_SOCK_CGROUP_DATA */
  740. #endif /* _LINUX_CGROUP_DEFS_H */